Background. Bone and joint infections (BJI) are severe infections that require a tailored and protracted antibiotic treatment. The diagnostic of BJI relies on the culture of surgical specimens, yet some bacteria would not grow because of extreme oxygen sensitivity or fastidious growth. Hence, metagenomic sequencing could potentially address those limitations. In this study, we assessed the performances of metagenomic sequencing of BJI samples for the identification of pathogens and the prediction of antibiotic susceptibility. Methods. A total of 179 samples were considered. The DNA was extracted with a kit aiming to decrease the amount of human DNA (Molzym), and sequenced on an Illumina HiSeq2500 in 2x250 paired-end reads. The taxonomy was obtained by MetaPhlAn2, the bacterial reads assembled with MetaSPAdes and the antibiotic resistance determinants (ARDs) identified using a database made of Resfinder+ARDs from functional metagenomic studies. Results. We could sequence the DNA from 24 out of 179 samples. For monomicrobial samples (n=8), the presence of the pathogen was confirmed by metagenomics in all cases. For polymicrobial samples (n=16), 32/55 bacteria (58.2%) were found at the species level (41/55 [74.5%] at the genus level). Conversely, a total of 273 bacteria not found in culture were identified, 182 being possible pathogens undetected in culture and 91 contaminants. A correct antibiotic susceptibility could be inferred in 94.1% cases for monomicrobial samples and in 76.5% cases in polymicrobial samples. Conclusions. When sufficient amounts of DNA can be extracted from samples, we found that clinical metagenomics is a potential tool to support conventional culture.